What happened?

I updated to v0.77.0 and it seems to introduce a regression with models on Opencode that don't expose different thinking modes. I have tested this with kimi-k2.6 on opencode go/zen and verified that i.e. glm-5.1 works fine (that model has different thinking levels).

Steps to reproduce

1. Connect to Opencode go/zen
2. use kimi-k2.6 as a model
3. type in any prompt
4. 400 Message gets shown

Expected behavior

pi should send a valid request to Opencode

Version

0.77.0